I'm surprised y'all stopped at the personal finance layer. I've been thinking for awhile that LLMs would be really effective as personal financial advisers, and this kind of hookup (plus I guess another one for investment accounts?) seems like all that's needed to bootstrap reasoning.
I recently did this for myself using https://tiller.com/ to sync checking/credit-card transactions to a Google Sheets spreadsheet. Then a GitHub action mirrors the spreadsheet to a free Supabase database.
From there, Supabase MCP or psql gives Claude/Codex access to the transactions/balances for english queries. Really impressed with their ability to find subscription patterns, abnormal patterns, etc. Also to predict cashflow which no online tool so far is good at i.e. "tell me how much cash I can move to savings based on my monthly spend patterns and available cash".
For autocategorization, I learned Claude is really good at custom DSLs. Had it create a markdown table based ruleset to normalize payee/categories. I also run the autocat rules as part of the GitHub actions.
Used driggsby the other day. Friend of mine sent it to me. Suprising how easily it found all of my subscriptions. That was the main reason i tried it. I'm sure i'll dive deeper into it this week though.
This seems like a solution looking for a problem. https://tiller.com/ works great and lets you do whatever calculations you want in a spreadsheet - and, bonus, it's never going to hallucinate.
I don't quite understand the desire to have these verbose summaries that you have to read from LLMs. You'll notice anomalies if you just categorize each of your expenses every so often (easy with Tiller).
Super cool writeup from a technology perspective. And a motivator to build more tools, play with MCP and try out Claude routines. The product pitch is also really clear. Good job all around.
Now for the criticism:
back in the day mint.com was mind-blowing. It's what you always thought you'd wanted. The graphs were interactive and pretty and you really loved seeing them go up. Not so much down. I was so attached to the gamified aspects, much like step counters. They reinforce habits.
My mint journey ended with roughly 5 years or so of data, once they sold to Intuit, didn't like the ads and willingly syncing all my data to mega-corp. Much like Duolingo, it felt good at the time, but I don't know that it did anything for me at all.
Tracking is a double-edged sword: it really does build better habits. It's better to track weight every day for example so you better understand that fluctuations are mostly noise. The daily tracking stuff is entirely useful to get the need to track daily out of your system.
TLDR: checking my net-worth daily sounds like something I should coach myself out from. Ironically that probably takes tools, but the end goal is to not need them.
Have to be careful with routines. There’s a very small disclaimer that’s barely noticeable that in routine mode all MCP tools, even write are always allowed. So agent can technically go rogue and start mutating your resources via MCP.
I don't really get this; could just use Plaid and have your own transaction/net worth site/app in a few hours with Claude Code and it'd be much more consistent.
My spouse was almost scammed out of 25k the other day, by a person. There is no way in fucking hell I'll let an "agent" touch my finances. That is madness.
To everyone who doesn’t know how Plaid works: You give your banking username and password directly to Plaid, and it keeps it (so it can continue to login).
I don’t understand how anyone is OK with this. It goes against every security principle and it’s against the terms and conditions of every bank.
I realize that almost no bank provides a secure and proper API to get info and/or to transfer funds, but Plaid’s solution is a disaster waiting to happen.
When I was trying to use Claude to analyze my past transactions, I found out that it was constantly hallucinating charges, sometimes adds new, double counts and etc.
When I'm dealing with my finances the 95% time Claude is correct and doesn't hallucinate is not enough as I have to be vigilant and review its work all the time. So it kinda makes it worthless in this case for me
In case anyone is interested, just wanted to share a few high level details about our infra/security setup.
- Backend & CLI are both strictly linted Rust. The webapp runs on Axum (Rust web framework), and connects to Postgres via sqlx.
- Financial read-only. There's no transfer, pay, or send tool in the product. Nothing in the AI surface can move money.
- We request transactions, investments, and liabilities from Plaid. We don't request auth, transfer, or payment_initiation, so we never receive full account numbers or routing numbers — just the last-4 mask Plaid returns by default.
- Bank usernames and passwords go to Plaid Link, not us. We only hold a per-institution access token.
- Plaid access tokens live in a separate database behind a single custody Cloud Run service, encrypted at rest by Cloud KMS. The broker calls KMS's encrypt/decrypt endpoints — the root key material never leaves Google's HSM boundary and the broker's service account is the only one with encrypt/decrypt permissions. The web app doesn't have permission to read that database.
- Every encryption and decryption call passes the Plaid item ID as AAD (additional authenticated data). A ciphertext from one item cannot be swapped in and decrypted as another item's token.
- Each Cloud Run service (including our web app) runs under its own cloud identity and with its own DB role.
- Internal calls between services are authenticated: the caller presents a short-lived identity token from the cloud provider, and the receiver verifies it.
- The prod databases have no public IP. Secrets live in managed secret storage, not in source or container images.
- The AI connector is OAuth 2.1 + PKCE, scoped per user, revocable from the UI. Every tool call records the tool name, sanitized args, calling client, and the reason the agent supplied, so you can see what your LLM asked on your behalf.
- There are no fetch-URL, shell, or general I/O tools in the AI surface. Tools return structured financial data and nothing else.
- Networking, IAM, and DB grants are all in Terraform. All infra changes go through that path.
Thank you so much for actually sharing these technical details. Shows you know THIS website’s audience.
And the thought put into security at every level gives me more confidence in the overall tool.
I’ve tried to build this myself (mvp for myself was just downloading statement PDFs manually and having Claude setting up ledger for plain text accounting, thinking I’ll eventually hookup plaid)
I’m mostly curious about how people work with plaid. Do you need a certain threshold of users to get started? Because I’d love to get a plaid account just for my finances. I’m not interested in building a product like this, just want to connect my personal and business accounts to a clean API.
Maybe my net worth is too low but I just don't see a value proposition. I don't want daily emails from LLMs and if I need updates on my investments any more often than quarterly (at most), I should probably seek safer investments. I am a bit interested in budgeting tools, but I want them to be completely deterministic. For me at least, financial planning is pretty uneventful and time spent optimizing expenses more than I already have would be better spent seeking a higher paying job.
I use actualbudget.org to track all spending, but only update investment accounts ("off-budget" in Actual Budget terms) once a month. Completely deterministic, as all things related to numbers should be.
I have pointed my LLM at the SQLite DB and asked it to tell me what it could see from my last five years of transactions, and I was impressed with the things it picked up, and what it reminded me of, but I'm not sure I saw any value in the sense of anything I would change.
I'm going to have it review things monthly to see if that helps me, but I'm not sure it will. I'm generally already aware of how my finances are going because of my budget updates.
We (Era Finance) have a solution for exactly this - Era Context, an MCP for connecting any compatible agent to your personal finances - https://era.app
It's focussed on read tools at the moment, but we have write tools coming (money transfers, debt paydown, etc)
Check it out and let me know if you'd like to see any particular features landing - just email alex at the above domain name.
For background context I'm Alex the CEO, I'm ~new to HN but was the former head of web presence at stripe.com and was at Square / CashApp before that. Hi!
Problem, I don’t trust 3rd party services with access to financial data. No offense to the OP.
I wish this was open source and self-hostable. Not suggesting that OP does that since I’m sure he has to pay bills :)
I may look into doing something like this for personal use. I’ve gotten interested in creating desktop apps which store data locally. Have created a few using Claude code. It seems Plaid has some APIs for personal use.
Yes, give it a try! I did this on my own first before deciding to make it "a thing." It's actually a really fun project and gives you a ton of insight into your finances. I'd recommend it.
Maybe the battle has long been lost, but why on earth would you want to entrust your entire financial transactions to an LLM? I bet LLM providers have even less safeguards regarding the use of these data than the financial industry, which to be clear is an abusive industry collecting, mining, and selling our data.
The main reason I do it, at least, is because the insights it provides (as someone who's interested in spending patterns/investing/etc.) are actually quite useful. With even a very basic prompt, I've found things that were otherwise completely missed.
Making it secure is actually quite hard, but that's why we're spending an enormous amount of time thinking about that.
My primary bank, Monzo (UK) provides a full API and trigger webhook for events.
This has enabled me to make a WhatsApp bot that asks me to explain abnormal transactions (using an LLM for reasoning), and also have a pre-midnight sweep balance to savings account to maximise interest daily interest.
I also maintain a small daily balance, and if I spend money during the day it tops it up from the savings to maintain that low balance. If I need to spend more, I can move money manually.
I was in this same situation a couple of months ago and I just sat down with my friend who is now my co-founder to https://neospend.com we both were having this issues and now we don’t anymore. Check it out.
34 comments
[ 0.26 ms ] story [ 50.2 ms ] threadThey have an api and I got llm to write a CLI for it.
That way agent can pretty much pull the data it needs or wants.
One thing I also had it do is build up a series of rules for tagging which then I run a cron for once a day.
Every once in a while I just ask it to look over the rules and make new ones for uncategorized transactions.
(As a side note I think having LLM “memoize” a task through a rule engine or code is really nice pattern)
But once you have the cli with query you can pretty much ask the agent to do anything.
From there, Supabase MCP or psql gives Claude/Codex access to the transactions/balances for english queries. Really impressed with their ability to find subscription patterns, abnormal patterns, etc. Also to predict cashflow which no online tool so far is good at i.e. "tell me how much cash I can move to savings based on my monthly spend patterns and available cash".
For autocategorization, I learned Claude is really good at custom DSLs. Had it create a markdown table based ruleset to normalize payee/categories. I also run the autocat rules as part of the GitHub actions.
I don't quite understand the desire to have these verbose summaries that you have to read from LLMs. You'll notice anomalies if you just categorize each of your expenses every so often (easy with Tiller).
Now for the criticism: back in the day mint.com was mind-blowing. It's what you always thought you'd wanted. The graphs were interactive and pretty and you really loved seeing them go up. Not so much down. I was so attached to the gamified aspects, much like step counters. They reinforce habits.
My mint journey ended with roughly 5 years or so of data, once they sold to Intuit, didn't like the ads and willingly syncing all my data to mega-corp. Much like Duolingo, it felt good at the time, but I don't know that it did anything for me at all.
Tracking is a double-edged sword: it really does build better habits. It's better to track weight every day for example so you better understand that fluctuations are mostly noise. The daily tracking stuff is entirely useful to get the need to track daily out of your system.
TLDR: checking my net-worth daily sounds like something I should coach myself out from. Ironically that probably takes tools, but the end goal is to not need them.
I don’t understand how anyone is OK with this. It goes against every security principle and it’s against the terms and conditions of every bank.
I realize that almost no bank provides a secure and proper API to get info and/or to transfer funds, but Plaid’s solution is a disaster waiting to happen.
When I'm dealing with my finances the 95% time Claude is correct and doesn't hallucinate is not enough as I have to be vigilant and review its work all the time. So it kinda makes it worthless in this case for me
- Backend & CLI are both strictly linted Rust. The webapp runs on Axum (Rust web framework), and connects to Postgres via sqlx.
- Financial read-only. There's no transfer, pay, or send tool in the product. Nothing in the AI surface can move money.
- We request transactions, investments, and liabilities from Plaid. We don't request auth, transfer, or payment_initiation, so we never receive full account numbers or routing numbers — just the last-4 mask Plaid returns by default.
- Bank usernames and passwords go to Plaid Link, not us. We only hold a per-institution access token.
- Plaid access tokens live in a separate database behind a single custody Cloud Run service, encrypted at rest by Cloud KMS. The broker calls KMS's encrypt/decrypt endpoints — the root key material never leaves Google's HSM boundary and the broker's service account is the only one with encrypt/decrypt permissions. The web app doesn't have permission to read that database.
- Every encryption and decryption call passes the Plaid item ID as AAD (additional authenticated data). A ciphertext from one item cannot be swapped in and decrypted as another item's token.
- Each Cloud Run service (including our web app) runs under its own cloud identity and with its own DB role.
- Internal calls between services are authenticated: the caller presents a short-lived identity token from the cloud provider, and the receiver verifies it.
- The prod databases have no public IP. Secrets live in managed secret storage, not in source or container images.
- The AI connector is OAuth 2.1 + PKCE, scoped per user, revocable from the UI. Every tool call records the tool name, sanitized args, calling client, and the reason the agent supplied, so you can see what your LLM asked on your behalf.
- There are no fetch-URL, shell, or general I/O tools in the AI surface. Tools return structured financial data and nothing else.
- Networking, IAM, and DB grants are all in Terraform. All infra changes go through that path.
- Infra access is gated by 2fa and security keys.
And the thought put into security at every level gives me more confidence in the overall tool.
I’ve tried to build this myself (mvp for myself was just downloading statement PDFs manually and having Claude setting up ledger for plain text accounting, thinking I’ll eventually hookup plaid)
I’m mostly curious about how people work with plaid. Do you need a certain threshold of users to get started? Because I’d love to get a plaid account just for my finances. I’m not interested in building a product like this, just want to connect my personal and business accounts to a clean API.
I have pointed my LLM at the SQLite DB and asked it to tell me what it could see from my last five years of transactions, and I was impressed with the things it picked up, and what it reminded me of, but I'm not sure I saw any value in the sense of anything I would change.
I'm going to have it review things monthly to see if that helps me, but I'm not sure it will. I'm generally already aware of how my finances are going because of my budget updates.
It's focussed on read tools at the moment, but we have write tools coming (money transfers, debt paydown, etc)
Check it out and let me know if you'd like to see any particular features landing - just email alex at the above domain name.
For background context I'm Alex the CEO, I'm ~new to HN but was the former head of web presence at stripe.com and was at Square / CashApp before that. Hi!
Problem, I don’t trust 3rd party services with access to financial data. No offense to the OP.
I wish this was open source and self-hostable. Not suggesting that OP does that since I’m sure he has to pay bills :)
I may look into doing something like this for personal use. I’ve gotten interested in creating desktop apps which store data locally. Have created a few using Claude code. It seems Plaid has some APIs for personal use.
So what’s the problem?
Making it secure is actually quite hard, but that's why we're spending an enormous amount of time thinking about that.
This has enabled me to make a WhatsApp bot that asks me to explain abnormal transactions (using an LLM for reasoning), and also have a pre-midnight sweep balance to savings account to maximise interest daily interest.
I also maintain a small daily balance, and if I spend money during the day it tops it up from the savings to maintain that low balance. If I need to spend more, I can move money manually.